SlideShare une entreprise Scribd logo
1  sur  25
Télécharger pour lire hors ligne
Geometry of Continuous-Time
Markov Chains
Shuchang Zhang
Content
› Motivation
› Mathematical Background
› Geometric Flow of Markov Chain
› Conclusions
Motivation
Motivation
• Time reversible Markov chain (detailed balance) is known to have
symmetric probability flux and can be described by gradient system
• Symmetric flux contributes to the production of relative entropy,
whereas skew-symmetric flux doesn’t
• Skew-symmetric flux playing a very important role as circulation in
time evolution of chains is yet hardly understood
• Evolution of Markov chain can be characterized by differential
geometry, which is a powerful and indispensable tool in dynamics
Mathematical Background
Alpha representation
In information geometry, a probability distribution can be coded by a
parameter 𝛼 as the following,
𝑙 𝛼
=
2
1 − 𝛼
𝑝
1−𝛼
2
Important examples include:
𝛼 = −1, 𝑙(−1) = 𝑝 mixed representation
𝛼 = 1, 𝑙(1)
= log 𝑝 exponential representation
𝛼 = 0, 𝑙(0) = 2 𝑝 0-representation
Alpha representation
𝛼 = −1 𝛼 = 1
Alpha representation
𝛼 = −1 𝛼 = 0
Alpha representation
• Different representations are equipped with different geometric
structures and restrict the dynamics of Markov chains on different
manifolds.
• Particularly, 0-representation admits the flow of probability on the
(hyper)sphere, which has radical symmetry.
Lie group of 𝑆𝑂 𝑛
• The motion on the manifold of 𝑛-sphere 𝑀 = 𝕊 𝑛 can be seen as
continuous isometry (distance-preserving) transformation.
• Given an initial point 𝑝0 ∈ 𝑀, the trajectory of 𝑝 is given by 𝑝𝑡 = 𝑔𝑡 𝑝0,
where 𝑔0 = 𝑒, 𝑔𝑡+𝑠 = 𝑔𝑡 𝑔𝑠 form a Lie group 𝐺 = 𝑆𝑂𝑛.
• Under matrix representation, 𝐺 is the set of order-𝑛 orthogonal
matrices with determinant 1. i.e. 𝑆𝑂𝑛 = 𝑂 ∈ 𝑆𝐿 𝑛|𝑂 𝑇
𝑂 = 𝑂𝑂 𝑇
= 𝐼 𝑛
Lie algebra of 𝔰𝔬 𝑛
• Let 𝐺 action on the torsor (principal homogenous space) 𝑀 from the
left, we have
ሶ𝑔𝑡 = lim
𝑠→0
𝑔𝑠 − 𝑒
𝑠
𝑔𝑡 = X𝑔𝑡 ∈ 𝑇𝑔 𝑡
𝐺
𝑋 = lim
𝑠→0
𝑔s − 𝑒
𝑠
= ሶ𝑔𝑡 ∘ 𝑔𝑡
−1
∈ 𝑇𝑒 𝐺 = 𝔤
The tangent vector X is the right translation of ሶ𝑔𝑡 by 𝑔𝑡
−1
.
• Lie algebra 𝔤 can be identified as tangent space at the identity.
Given any vector 𝑋 ∈ 𝔤, there is a unique left-invariant vector field
𝑋 𝑔 = 𝑇𝐿 𝑔 𝑋 = 𝐿 𝑔∗
𝑋
Lie algebra of 𝔰𝔬 𝑛
• Note that for 𝑂𝑠 ∈ 𝑆𝑂𝑛 near the identity, we have
𝐼 = 𝑂𝑠
𝑇 𝑂𝑠 = 𝐼 + 𝑠Ω + 𝑜 𝑠
𝑇
𝐼 + 𝑠Ω + 𝑜 𝑠 = 𝐼 + 𝑠 Ω + Ω 𝑇 + 𝑜 𝑠
The matrix Lie algebra of 𝔰𝔬 𝑛 is the set of skew-symmetric matrices,
i.e. 𝔰𝔬 𝑛 = 𝑇𝑒 𝑆𝑂𝑛 = Ω ∈ 𝐺𝐿 𝑛| Ω + Ω 𝑇
= 0
• It can also be identified with vector space of dimension
𝑛(𝑛−1)
2
Adjoint and coadjoint representation of 𝔰𝔬 𝑛
• An important representation of Lie algebra, called adjoint
representation, is defined as
𝑎𝑑 ∶ 𝔤 → 𝔤𝔩 𝑛 = 𝐸𝑛𝑑 𝔤
𝑎𝑑 𝑋: 𝑌 ↦ 𝑋, 𝑌
• Choose a non-degenerate inner product , on Lie algebra 𝔤, the
coadjoint representation is defined as
𝑎𝑑 𝑍
∗
𝑋, 𝑌 = 𝑋, 𝑎𝑑 𝑍 𝑌
Riemannian metric
• The inner product , induces a right-invariant Riemannian metric
, 𝑔 on the whole Lie group 𝐺. Given two vectors 𝑋, 𝑌 ∈ 𝑇𝑔 𝐺, the
Riemannian metric is defined as
𝑋, 𝑌 𝑔: = 𝑇𝑅 𝑔
−1
∗
𝑋 , 𝑇𝑅 𝑔
−1
∗
𝑌
• The geodesic is defined as the extremal of the energy functional
𝐸 𝑔𝑡 = න
𝑎
𝑏
1
2
ሶ𝑔𝑡, ሶ𝑔𝑡 𝑔 𝑑𝑡
Geometric Flow of Markov Chain
0-representation of Markov chain
• A continuous-time Markov chain (CTMC) is completely determined by
its infinitesimal generator 𝑄, admitting the first-order ODE.
ሶ𝑝𝑖 = ෍
𝑗
𝑄𝑖𝑗 𝑝𝑗
where σ𝑖 𝑄𝑖𝑗 = 0, 𝑄𝑖𝑗 ≥ 0 for 𝑖 ≠ 𝑗 and 𝑄𝑖𝑖 < 0
• Let 𝑞𝑖 = 2 𝑝𝑖 be 0-representation of probability 𝑝, we have
ሶ𝑞𝑖 =
1
2
෍
𝑗
𝑄𝑖𝑗 𝑞 𝑗
2
𝑞𝑖
= ෍
𝑗
Ω𝑖𝑗 𝑞 𝑗
where Ω𝑖𝑗 + Ω𝑗𝑖 = 0
Evolution of the same CTMC
𝛼 = −1 𝛼 = 0
Geometric flow of CTMC
• Let 𝑞𝑡 be a continuous trajectory on 𝕊 𝑛 such that 𝑞𝑡 = 𝑔𝑡 𝑞0, where
𝑔𝑡 ∈ 𝑆𝑂𝑛, then
ሶ𝑞𝑡 = ሶ𝑔𝑡 𝑞0 = ሶ𝑔𝑡 𝑔𝑡
−1
𝑞𝑡 = Ω𝑞𝑡
Ω = ሶ𝑔𝑡 𝑔𝑡
−1
∈ 𝔰𝔬 𝑛
• This establishes a bijection between the trajectory on 𝕊 𝑛 and that on
𝑆𝑂𝑛. This inspires us to investigate geodesic flow on 𝑆𝑂𝑛.
Geodesic flow on 𝑆𝑂 𝑛
• By requiring the first variation of energy functional 𝐸[𝑔𝑡] to vanish,
i.e. we have
𝛿𝐸 𝑔𝑡 =
1
2
𝛿 න
𝑎
𝑏
ሶ𝑔𝑡, ሶ𝑔𝑡 𝑔 𝑑𝑡 =
1
2
𝛿 න
𝑎
𝑏
ሶ𝑔𝑡 𝑔𝑡
−1
, ሶ𝑔𝑡 𝑔𝑡
−1
𝑑𝑡
= න
𝑎
𝑏
𝛿 ሶ𝑔𝑡 𝑔𝑡
−1
+ ሶ𝑔𝑡 𝛿𝑔𝑡
−1
, ሶ𝑔𝑡 𝑔𝑡
−1
𝑑𝑡 = න
𝑎
𝑏
ሶ𝛿𝑔𝑡 𝑔𝑡
−1
− Ω𝛿𝑔𝑡 𝑔𝑡
−1
, Ω 𝑑𝑡
= 𝛿𝑔𝑡 𝑔𝑡
−1
, Ω ቚ
𝑎
𝑏
+ න
𝑎
𝑏
𝛿𝑔𝑔𝑡
−1
, Ω + 𝛿𝑔𝑡 𝑔𝑡
−1 ሶΩ, Ω 𝑑𝑡
= න
𝑎
𝑏
𝛿𝑔𝑡 𝑔𝑡
−1 ሶΩ − 𝑎𝑑Ω 𝛿𝑔𝑔𝑡
−1
, Ω 𝑑𝑡 = න
𝑎
𝑏
𝛿𝑔𝑔𝑡
−1
, ሶΩ − 𝑎𝑑Ω
∗
Ω 𝑑𝑡 = 0
Geodesic flow on 𝑆𝑂 𝑛
• We obtain Euler-Poincare equation
ሶΩ = 𝑎𝑑Ω
∗
Ω
• Choose Frobenius inner product 𝑋, 𝑌 = 𝑡𝑟(𝑋 𝑇 𝑌), then
𝑋, 𝑎𝑑 𝑍 𝑌 = 𝑋, 𝑍, 𝑌 = 𝑡𝑟 𝑋 𝑇 𝑍𝑌 − 𝑌𝑍
= 𝑡𝑟 𝑋 𝑇 𝑍 − 𝑍𝑋 𝑇 𝑌 = 𝑍 𝑇, 𝑋 , 𝑌 = 𝑎𝑑 𝑍
∗
𝑋, 𝑌
• Rewrite Euler-Poincare equation as Lie-Poisson form
ሶΩ + Ω, Ω = 0
Geometric flow of CTMC again
• Euler-Poincare equation:
ሶΩ = 𝑎𝑑Ω
∗
Ω
Note that this equation doesn’t contain 𝑔𝑡 explicitly.
• We can reconstruct the equation of the motion of 0-representation
Markov chain by
Ω = ሶ𝑔𝑡 𝑔𝑡
−1
,
ሶ𝑔𝑡 = Ω𝑔𝑡
Conservation law in CTMC
• By Noether’s theorem, the right-invariant geodesic flow preserves
some quantities, which can be computed by momentum map 𝜇
𝜇: 𝔤 → ℝ, 𝑋 ↦ Ω, 𝑔𝑡 𝑋𝑔𝑡
−1
• Proof
ሶ𝜇 = ሶΩ, 𝑔𝑡 𝑋𝑔𝑡
−1
+ Ω, Ω, 𝑔𝑡 𝑋𝑔𝑡
−1
= 𝑎𝑑ΩΩ, 𝑔𝑡 𝑋𝑔𝑡
−1
+ Ω, 𝑎𝑑Ω 𝑔𝑡 𝑋𝑔𝑡
−1
= 𝑎𝑑ΩΩ, 𝑔𝑡 𝑋𝑔𝑡
−1
+ 𝑎𝑑Ω
∗
𝛺, 𝑔𝑡 𝑋𝑔𝑡
−1
= 0
Conclusions
Conclusions
In summary, we give a geometric formulation of 0-representation CTMC.
This view allows us to
• Investigate the dynamics on (hyper)sphere, from both intrinsic
and extrinsic view
• Reduce the dimension of infinitesimal generator by half (from
𝑛(𝑛 − 1) to
𝑛 𝑛−1
2
)
• The time evolution of Markov chains follows Euler-Poincare
equation, whose trajectory is always geodesic flow
• Conservation quantities can be found
Further questions
There are many problems to be solved yet
• How to distinguish skew-symmetric flux from symmetric one in
geometric view
• Geometric formulation of CTMC in other representations
• Find master equation of geodesic flows
• Etc..

Contenu connexe

Tendances

Lagrange's Theorem
Lagrange's TheoremLagrange's Theorem
Lagrange's Theorem
john1129
 
Operators n dirac in qm
Operators n dirac in qmOperators n dirac in qm
Operators n dirac in qm
Anda Tywabi
 
Cp project-1-chaotic-pendulum
Cp project-1-chaotic-pendulumCp project-1-chaotic-pendulum
Cp project-1-chaotic-pendulum
Reynan Toledo
 
REU Research Project_Final
REU Research Project_FinalREU Research Project_Final
REU Research Project_Final
Jongyoon Sohn
 

Tendances (20)

ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02
 
Me314 week09-root locusanalysis
Me314 week09-root locusanalysisMe314 week09-root locusanalysis
Me314 week09-root locusanalysis
 
Chemical Bonding
Chemical BondingChemical Bonding
Chemical Bonding
 
The time independent Schrödinger wave equation
The time independent Schrödinger wave equationThe time independent Schrödinger wave equation
The time independent Schrödinger wave equation
 
Mathematical Formulation of Quantum Mechanics
Mathematical Formulation of Quantum Mechanics Mathematical Formulation of Quantum Mechanics
Mathematical Formulation of Quantum Mechanics
 
Lagrange's Theorem
Lagrange's TheoremLagrange's Theorem
Lagrange's Theorem
 
Non-linear control of a bipedal (Three-Linked) Walker using feedback Lineariz...
Non-linear control of a bipedal (Three-Linked) Walker using feedback Lineariz...Non-linear control of a bipedal (Three-Linked) Walker using feedback Lineariz...
Non-linear control of a bipedal (Three-Linked) Walker using feedback Lineariz...
 
Operators n dirac in qm
Operators n dirac in qmOperators n dirac in qm
Operators n dirac in qm
 
Simple harmonic oscillator - Classical Mechanics
Simple harmonic oscillator - Classical MechanicsSimple harmonic oscillator - Classical Mechanics
Simple harmonic oscillator - Classical Mechanics
 
Robust Fuzzy Output Feedback Controller for Affine Nonlinear Systems via T–S ...
Robust Fuzzy Output Feedback Controller for Affine Nonlinear Systems via T–S ...Robust Fuzzy Output Feedback Controller for Affine Nonlinear Systems via T–S ...
Robust Fuzzy Output Feedback Controller for Affine Nonlinear Systems via T–S ...
 
Energy-Based Control of Under-Actuated Mechanical Systems - Remotely Driven A...
Energy-Based Control of Under-Actuated Mechanical Systems - Remotely Driven A...Energy-Based Control of Under-Actuated Mechanical Systems - Remotely Driven A...
Energy-Based Control of Under-Actuated Mechanical Systems - Remotely Driven A...
 
Spherical harmonics
Spherical harmonicsSpherical harmonics
Spherical harmonics
 
NONLINEAR CONTROL SYSTEM (Phase plane & Phase Trajectory Method)
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method)NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method)
NONLINEAR CONTROL SYSTEM (Phase plane & Phase Trajectory Method)
 
Postulates of quantum mechanics & operators
Postulates of quantum mechanics & operatorsPostulates of quantum mechanics & operators
Postulates of quantum mechanics & operators
 
Quick run through on classical mechancis and quantum mechanics
Quick run through on classical mechancis and quantum mechanics Quick run through on classical mechancis and quantum mechanics
Quick run through on classical mechancis and quantum mechanics
 
Cp project-1-chaotic-pendulum
Cp project-1-chaotic-pendulumCp project-1-chaotic-pendulum
Cp project-1-chaotic-pendulum
 
Simrock 3
Simrock 3Simrock 3
Simrock 3
 
Damped harmonic oscillator
Damped harmonic oscillatorDamped harmonic oscillator
Damped harmonic oscillator
 
3 analytical kinematics
3 analytical kinematics3 analytical kinematics
3 analytical kinematics
 
REU Research Project_Final
REU Research Project_FinalREU Research Project_Final
REU Research Project_Final
 

Similaire à Geometry of Continuous Time Markov Chains

Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucksFully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
Joseph Hucks, Ph.D.
 
Max flows via electrical flows (long talk)
Max flows via electrical flows (long talk)Max flows via electrical flows (long talk)
Max flows via electrical flows (long talk)
Thatchaphol Saranurak
 
Perturbation
PerturbationPerturbation
Perturbation
BHAVANAR12
 
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdfNONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
AliMaarouf5
 

Similaire à Geometry of Continuous Time Markov Chains (20)

Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucksFully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
Fully-Polarimetric_Phased_Array_Far_Field_Modeling_2015SECONWorkshop_JHucks
 
Max flows via electrical flows (long talk)
Max flows via electrical flows (long talk)Max flows via electrical flows (long talk)
Max flows via electrical flows (long talk)
 
The wkb approximation..
The wkb approximation..The wkb approximation..
The wkb approximation..
 
The wkb approximation
The wkb approximationThe wkb approximation
The wkb approximation
 
Small amplitude oscillations
Small amplitude oscillationsSmall amplitude oscillations
Small amplitude oscillations
 
Optimum Engineering Design - Day 2b. Classical Optimization methods
Optimum Engineering Design - Day 2b. Classical Optimization methodsOptimum Engineering Design - Day 2b. Classical Optimization methods
Optimum Engineering Design - Day 2b. Classical Optimization methods
 
Perturbation
PerturbationPerturbation
Perturbation
 
Gauge Theory for Beginners.pptx
Gauge Theory for Beginners.pptxGauge Theory for Beginners.pptx
Gauge Theory for Beginners.pptx
 
String theory basics
String theory basicsString theory basics
String theory basics
 
Navier stokes equation in coordinates binormal, tangent and normal
Navier stokes equation in coordinates binormal, tangent and normalNavier stokes equation in coordinates binormal, tangent and normal
Navier stokes equation in coordinates binormal, tangent and normal
 
Orbital_Simulation (2).pptx
Orbital_Simulation (2).pptxOrbital_Simulation (2).pptx
Orbital_Simulation (2).pptx
 
Relativistic formulation of Maxwell equations.
Relativistic formulation of Maxwell equations.Relativistic formulation of Maxwell equations.
Relativistic formulation of Maxwell equations.
 
Probabilistic Models of Time Series and Sequences
Probabilistic Models of Time Series and SequencesProbabilistic Models of Time Series and Sequences
Probabilistic Models of Time Series and Sequences
 
Basic calculus (i)
Basic calculus (i)Basic calculus (i)
Basic calculus (i)
 
Magnetic Monopoles, Duality and SUSY.pptx
Magnetic Monopoles, Duality and SUSY.pptxMagnetic Monopoles, Duality and SUSY.pptx
Magnetic Monopoles, Duality and SUSY.pptx
 
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdfNONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
NONLINEAR CONTROL SYSTEM(Phase plane & Phase Trajectory Method).pdf
 
Module 1 Theory of Matrices.pdf
Module 1 Theory of Matrices.pdfModule 1 Theory of Matrices.pdf
Module 1 Theory of Matrices.pdf
 
Lecture Notes: EEEC4340318 Instrumentation and Control Systems - System Models
Lecture Notes:  EEEC4340318 Instrumentation and Control Systems - System ModelsLecture Notes:  EEEC4340318 Instrumentation and Control Systems - System Models
Lecture Notes: EEEC4340318 Instrumentation and Control Systems - System Models
 
COORDINATE SYSTEM.pdf
COORDINATE SYSTEM.pdfCOORDINATE SYSTEM.pdf
COORDINATE SYSTEM.pdf
 
COORDINATE SYSTEM.pdf
COORDINATE SYSTEM.pdfCOORDINATE SYSTEM.pdf
COORDINATE SYSTEM.pdf
 

Dernier

Formation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disksFormation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disks
Sérgio Sacani
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
PirithiRaju
 
Seismic Method Estimate velocity from seismic data.pptx
Seismic Method Estimate velocity from seismic  data.pptxSeismic Method Estimate velocity from seismic  data.pptx
Seismic Method Estimate velocity from seismic data.pptx
AlMamun560346
 
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune WaterworldsBiogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
Sérgio Sacani
 
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
ssuser79fe74
 

Dernier (20)

❤Jammu Kashmir Call Girls 8617697112 Personal Whatsapp Number 💦✅.
❤Jammu Kashmir Call Girls 8617697112 Personal Whatsapp Number 💦✅.❤Jammu Kashmir Call Girls 8617697112 Personal Whatsapp Number 💦✅.
❤Jammu Kashmir Call Girls 8617697112 Personal Whatsapp Number 💦✅.
 
Feature-aligned N-BEATS with Sinkhorn divergence (ICLR '24)
Feature-aligned N-BEATS with Sinkhorn divergence (ICLR '24)Feature-aligned N-BEATS with Sinkhorn divergence (ICLR '24)
Feature-aligned N-BEATS with Sinkhorn divergence (ICLR '24)
 
Formation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disksFormation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disks
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
 
SAMASTIPUR CALL GIRL 7857803690 LOW PRICE ESCORT SERVICE
SAMASTIPUR CALL GIRL 7857803690  LOW PRICE  ESCORT SERVICESAMASTIPUR CALL GIRL 7857803690  LOW PRICE  ESCORT SERVICE
SAMASTIPUR CALL GIRL 7857803690 LOW PRICE ESCORT SERVICE
 
Seismic Method Estimate velocity from seismic data.pptx
Seismic Method Estimate velocity from seismic  data.pptxSeismic Method Estimate velocity from seismic  data.pptx
Seismic Method Estimate velocity from seismic data.pptx
 
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune WaterworldsBiogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
Biogenic Sulfur Gases as Biosignatures on Temperate Sub-Neptune Waterworlds
 
Nanoparticles synthesis and characterization​ ​
Nanoparticles synthesis and characterization​  ​Nanoparticles synthesis and characterization​  ​
Nanoparticles synthesis and characterization​ ​
 
FAIRSpectra - Enabling the FAIRification of Spectroscopy and Spectrometry
FAIRSpectra - Enabling the FAIRification of Spectroscopy and SpectrometryFAIRSpectra - Enabling the FAIRification of Spectroscopy and Spectrometry
FAIRSpectra - Enabling the FAIRification of Spectroscopy and Spectrometry
 
GBSN - Biochemistry (Unit 1)
GBSN - Biochemistry (Unit 1)GBSN - Biochemistry (Unit 1)
GBSN - Biochemistry (Unit 1)
 
Hire 💕 9907093804 Hooghly Call Girls Service Call Girls Agency
Hire 💕 9907093804 Hooghly Call Girls Service Call Girls AgencyHire 💕 9907093804 Hooghly Call Girls Service Call Girls Agency
Hire 💕 9907093804 Hooghly Call Girls Service Call Girls Agency
 
Chemistry 4th semester series (krishna).pdf
Chemistry 4th semester series (krishna).pdfChemistry 4th semester series (krishna).pdf
Chemistry 4th semester series (krishna).pdf
 
GBSN - Microbiology (Unit 3)
GBSN - Microbiology (Unit 3)GBSN - Microbiology (Unit 3)
GBSN - Microbiology (Unit 3)
 
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
 
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
 
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRLKochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
 
9654467111 Call Girls In Raj Nagar Delhi Short 1500 Night 6000
9654467111 Call Girls In Raj Nagar Delhi Short 1500 Night 60009654467111 Call Girls In Raj Nagar Delhi Short 1500 Night 6000
9654467111 Call Girls In Raj Nagar Delhi Short 1500 Night 6000
 
Forensic Biology & Its biological significance.pdf
Forensic Biology & Its biological significance.pdfForensic Biology & Its biological significance.pdf
Forensic Biology & Its biological significance.pdf
 
module for grade 9 for distance learning
module for grade 9 for distance learningmodule for grade 9 for distance learning
module for grade 9 for distance learning
 
Botany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfBotany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdf
 

Geometry of Continuous Time Markov Chains

  • 1. Geometry of Continuous-Time Markov Chains Shuchang Zhang
  • 2. Content › Motivation › Mathematical Background › Geometric Flow of Markov Chain › Conclusions
  • 4. Motivation • Time reversible Markov chain (detailed balance) is known to have symmetric probability flux and can be described by gradient system • Symmetric flux contributes to the production of relative entropy, whereas skew-symmetric flux doesn’t • Skew-symmetric flux playing a very important role as circulation in time evolution of chains is yet hardly understood • Evolution of Markov chain can be characterized by differential geometry, which is a powerful and indispensable tool in dynamics
  • 6. Alpha representation In information geometry, a probability distribution can be coded by a parameter 𝛼 as the following, 𝑙 𝛼 = 2 1 − 𝛼 𝑝 1−𝛼 2 Important examples include: 𝛼 = −1, 𝑙(−1) = 𝑝 mixed representation 𝛼 = 1, 𝑙(1) = log 𝑝 exponential representation 𝛼 = 0, 𝑙(0) = 2 𝑝 0-representation
  • 9. Alpha representation • Different representations are equipped with different geometric structures and restrict the dynamics of Markov chains on different manifolds. • Particularly, 0-representation admits the flow of probability on the (hyper)sphere, which has radical symmetry.
  • 10. Lie group of 𝑆𝑂 𝑛 • The motion on the manifold of 𝑛-sphere 𝑀 = 𝕊 𝑛 can be seen as continuous isometry (distance-preserving) transformation. • Given an initial point 𝑝0 ∈ 𝑀, the trajectory of 𝑝 is given by 𝑝𝑡 = 𝑔𝑡 𝑝0, where 𝑔0 = 𝑒, 𝑔𝑡+𝑠 = 𝑔𝑡 𝑔𝑠 form a Lie group 𝐺 = 𝑆𝑂𝑛. • Under matrix representation, 𝐺 is the set of order-𝑛 orthogonal matrices with determinant 1. i.e. 𝑆𝑂𝑛 = 𝑂 ∈ 𝑆𝐿 𝑛|𝑂 𝑇 𝑂 = 𝑂𝑂 𝑇 = 𝐼 𝑛
  • 11. Lie algebra of 𝔰𝔬 𝑛 • Let 𝐺 action on the torsor (principal homogenous space) 𝑀 from the left, we have ሶ𝑔𝑡 = lim 𝑠→0 𝑔𝑠 − 𝑒 𝑠 𝑔𝑡 = X𝑔𝑡 ∈ 𝑇𝑔 𝑡 𝐺 𝑋 = lim 𝑠→0 𝑔s − 𝑒 𝑠 = ሶ𝑔𝑡 ∘ 𝑔𝑡 −1 ∈ 𝑇𝑒 𝐺 = 𝔤 The tangent vector X is the right translation of ሶ𝑔𝑡 by 𝑔𝑡 −1 . • Lie algebra 𝔤 can be identified as tangent space at the identity. Given any vector 𝑋 ∈ 𝔤, there is a unique left-invariant vector field 𝑋 𝑔 = 𝑇𝐿 𝑔 𝑋 = 𝐿 𝑔∗ 𝑋
  • 12. Lie algebra of 𝔰𝔬 𝑛 • Note that for 𝑂𝑠 ∈ 𝑆𝑂𝑛 near the identity, we have 𝐼 = 𝑂𝑠 𝑇 𝑂𝑠 = 𝐼 + 𝑠Ω + 𝑜 𝑠 𝑇 𝐼 + 𝑠Ω + 𝑜 𝑠 = 𝐼 + 𝑠 Ω + Ω 𝑇 + 𝑜 𝑠 The matrix Lie algebra of 𝔰𝔬 𝑛 is the set of skew-symmetric matrices, i.e. 𝔰𝔬 𝑛 = 𝑇𝑒 𝑆𝑂𝑛 = Ω ∈ 𝐺𝐿 𝑛| Ω + Ω 𝑇 = 0 • It can also be identified with vector space of dimension 𝑛(𝑛−1) 2
  • 13. Adjoint and coadjoint representation of 𝔰𝔬 𝑛 • An important representation of Lie algebra, called adjoint representation, is defined as 𝑎𝑑 ∶ 𝔤 → 𝔤𝔩 𝑛 = 𝐸𝑛𝑑 𝔤 𝑎𝑑 𝑋: 𝑌 ↦ 𝑋, 𝑌 • Choose a non-degenerate inner product , on Lie algebra 𝔤, the coadjoint representation is defined as 𝑎𝑑 𝑍 ∗ 𝑋, 𝑌 = 𝑋, 𝑎𝑑 𝑍 𝑌
  • 14. Riemannian metric • The inner product , induces a right-invariant Riemannian metric , 𝑔 on the whole Lie group 𝐺. Given two vectors 𝑋, 𝑌 ∈ 𝑇𝑔 𝐺, the Riemannian metric is defined as 𝑋, 𝑌 𝑔: = 𝑇𝑅 𝑔 −1 ∗ 𝑋 , 𝑇𝑅 𝑔 −1 ∗ 𝑌 • The geodesic is defined as the extremal of the energy functional 𝐸 𝑔𝑡 = න 𝑎 𝑏 1 2 ሶ𝑔𝑡, ሶ𝑔𝑡 𝑔 𝑑𝑡
  • 15. Geometric Flow of Markov Chain
  • 16. 0-representation of Markov chain • A continuous-time Markov chain (CTMC) is completely determined by its infinitesimal generator 𝑄, admitting the first-order ODE. ሶ𝑝𝑖 = ෍ 𝑗 𝑄𝑖𝑗 𝑝𝑗 where σ𝑖 𝑄𝑖𝑗 = 0, 𝑄𝑖𝑗 ≥ 0 for 𝑖 ≠ 𝑗 and 𝑄𝑖𝑖 < 0 • Let 𝑞𝑖 = 2 𝑝𝑖 be 0-representation of probability 𝑝, we have ሶ𝑞𝑖 = 1 2 ෍ 𝑗 𝑄𝑖𝑗 𝑞 𝑗 2 𝑞𝑖 = ෍ 𝑗 Ω𝑖𝑗 𝑞 𝑗 where Ω𝑖𝑗 + Ω𝑗𝑖 = 0
  • 17. Evolution of the same CTMC 𝛼 = −1 𝛼 = 0
  • 18. Geometric flow of CTMC • Let 𝑞𝑡 be a continuous trajectory on 𝕊 𝑛 such that 𝑞𝑡 = 𝑔𝑡 𝑞0, where 𝑔𝑡 ∈ 𝑆𝑂𝑛, then ሶ𝑞𝑡 = ሶ𝑔𝑡 𝑞0 = ሶ𝑔𝑡 𝑔𝑡 −1 𝑞𝑡 = Ω𝑞𝑡 Ω = ሶ𝑔𝑡 𝑔𝑡 −1 ∈ 𝔰𝔬 𝑛 • This establishes a bijection between the trajectory on 𝕊 𝑛 and that on 𝑆𝑂𝑛. This inspires us to investigate geodesic flow on 𝑆𝑂𝑛.
  • 19. Geodesic flow on 𝑆𝑂 𝑛 • By requiring the first variation of energy functional 𝐸[𝑔𝑡] to vanish, i.e. we have 𝛿𝐸 𝑔𝑡 = 1 2 𝛿 න 𝑎 𝑏 ሶ𝑔𝑡, ሶ𝑔𝑡 𝑔 𝑑𝑡 = 1 2 𝛿 න 𝑎 𝑏 ሶ𝑔𝑡 𝑔𝑡 −1 , ሶ𝑔𝑡 𝑔𝑡 −1 𝑑𝑡 = න 𝑎 𝑏 𝛿 ሶ𝑔𝑡 𝑔𝑡 −1 + ሶ𝑔𝑡 𝛿𝑔𝑡 −1 , ሶ𝑔𝑡 𝑔𝑡 −1 𝑑𝑡 = න 𝑎 𝑏 ሶ𝛿𝑔𝑡 𝑔𝑡 −1 − Ω𝛿𝑔𝑡 𝑔𝑡 −1 , Ω 𝑑𝑡 = 𝛿𝑔𝑡 𝑔𝑡 −1 , Ω ቚ 𝑎 𝑏 + න 𝑎 𝑏 𝛿𝑔𝑔𝑡 −1 , Ω + 𝛿𝑔𝑡 𝑔𝑡 −1 ሶΩ, Ω 𝑑𝑡 = න 𝑎 𝑏 𝛿𝑔𝑡 𝑔𝑡 −1 ሶΩ − 𝑎𝑑Ω 𝛿𝑔𝑔𝑡 −1 , Ω 𝑑𝑡 = න 𝑎 𝑏 𝛿𝑔𝑔𝑡 −1 , ሶΩ − 𝑎𝑑Ω ∗ Ω 𝑑𝑡 = 0
  • 20. Geodesic flow on 𝑆𝑂 𝑛 • We obtain Euler-Poincare equation ሶΩ = 𝑎𝑑Ω ∗ Ω • Choose Frobenius inner product 𝑋, 𝑌 = 𝑡𝑟(𝑋 𝑇 𝑌), then 𝑋, 𝑎𝑑 𝑍 𝑌 = 𝑋, 𝑍, 𝑌 = 𝑡𝑟 𝑋 𝑇 𝑍𝑌 − 𝑌𝑍 = 𝑡𝑟 𝑋 𝑇 𝑍 − 𝑍𝑋 𝑇 𝑌 = 𝑍 𝑇, 𝑋 , 𝑌 = 𝑎𝑑 𝑍 ∗ 𝑋, 𝑌 • Rewrite Euler-Poincare equation as Lie-Poisson form ሶΩ + Ω, Ω = 0
  • 21. Geometric flow of CTMC again • Euler-Poincare equation: ሶΩ = 𝑎𝑑Ω ∗ Ω Note that this equation doesn’t contain 𝑔𝑡 explicitly. • We can reconstruct the equation of the motion of 0-representation Markov chain by Ω = ሶ𝑔𝑡 𝑔𝑡 −1 , ሶ𝑔𝑡 = Ω𝑔𝑡
  • 22. Conservation law in CTMC • By Noether’s theorem, the right-invariant geodesic flow preserves some quantities, which can be computed by momentum map 𝜇 𝜇: 𝔤 → ℝ, 𝑋 ↦ Ω, 𝑔𝑡 𝑋𝑔𝑡 −1 • Proof ሶ𝜇 = ሶΩ, 𝑔𝑡 𝑋𝑔𝑡 −1 + Ω, Ω, 𝑔𝑡 𝑋𝑔𝑡 −1 = 𝑎𝑑ΩΩ, 𝑔𝑡 𝑋𝑔𝑡 −1 + Ω, 𝑎𝑑Ω 𝑔𝑡 𝑋𝑔𝑡 −1 = 𝑎𝑑ΩΩ, 𝑔𝑡 𝑋𝑔𝑡 −1 + 𝑎𝑑Ω ∗ 𝛺, 𝑔𝑡 𝑋𝑔𝑡 −1 = 0
  • 24. Conclusions In summary, we give a geometric formulation of 0-representation CTMC. This view allows us to • Investigate the dynamics on (hyper)sphere, from both intrinsic and extrinsic view • Reduce the dimension of infinitesimal generator by half (from 𝑛(𝑛 − 1) to 𝑛 𝑛−1 2 ) • The time evolution of Markov chains follows Euler-Poincare equation, whose trajectory is always geodesic flow • Conservation quantities can be found
  • 25. Further questions There are many problems to be solved yet • How to distinguish skew-symmetric flux from symmetric one in geometric view • Geometric formulation of CTMC in other representations • Find master equation of geodesic flows • Etc..